Trouble while using a GPU safe random function

[attarnasir]

Hey AMReX folks,
I have a MultiFab that I want to fill with random numbers around the value 0.4. Using amrex::Random() seamlessly works on the CPU, but I'm getting an error for the GPU. The snippet of my code is as follows,

for (MFIter mfi(phi_new); mfi.isValid(); ++mfi)
    {
        const Box& vbx = mfi.validbox();
        auto const& phiNew = phi_new.array(mfi);
        amrex::ParallelFor(vbx,
        [=] AMREX_GPU_DEVICE(int i, int j, int k)
        {   
            Real noise = 0.02;
            amrex::RandomEngine rd;
            phiNew(i,j,k) = 0.4 + noise*(0.5 - amrex::Random(rd));

        });
    }

The code compiles when USE_CUDA is set to TRUE. But gives the following runtime error,

amrex::Abort::0::CUDA error 719 in file /home/amrex/Src/Base/AMReX_GpuDevice.cpp line 614: unspecified launch failure !!!
SIGABRT
/usr/bin/addr2line: Dwarf Error: found dwarf version '5', this reader only handles version 2, 3 and 4 information.
see Backtrace.0 file for details

I believe I'm using RandomEngine incorrectly here. If anyone knows how to solve this problem, please guide me. Any advice would be greatly appreciated!

Thank You

[WeiqunZhang]

Try

        amrex::ParallelForRNG(vbx,
        [=] AMREX_GPU_DEVICE(int i, int j, int k, RandomEngine const& rd)
        {   
            Real noise = 0.02;
            phiNew(i,j,k) = 0.4 + noise*(0.5 - amrex::Random(rd));
        });

[attarnasir]

Thank you for the help!